Capturing the contributions of the semantic web to the IoT: a unifying vision

Tracking #: 1644-2856

Nicolas Seydoux
Nathalie Hernandez
Khalil Drira
Thierry Monteil

Responsible editor: 
Guest Editors IoT 2017

Submission type: 
Survey Article
The Internet of Things (IoT) is a technological topic with a very important societal impact. IoT application domains are various and include: smart cities, precision farming, smart factories, and smart buildings. The diversity of these application domains is the source of the very high technological heterogeneity in the IoT, leading to interoperability issues. The semantic web principles and technologies are more and more adopted as a solution to these interoperability issues, leading to the emergence of a new domain, the Semantic Web Of Things (SWoT). Scientific contributions to the SWoT are many, and the diversity of architectures in which they are expressed complicates comparison. To unify the presented architectures, we propose an architectural pattern, LMU-N. LMU-N provides a reading grid used to classify processes to which the SWoT community contributes, and to describe how the semantic web impacts the IoT. Then, the evolutions of the semantic web to adapt to the IoT constraints are described as well, in order to give a twofold view of the convergence between the IoT and the semantic web toward the SWoT.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
Anonymous submitted on 11/Jul/2017
Major Revision
Review Comment:

The article surveys contributions of the semantic web to the internet of things. Authors propose a model, LMU-N to classify the contributions and analyse the current trends. LMU-N is a graph model, where nodes (partitioned in lower, medium and upper nodes) are connected through message flows. The article is clearly on topic with the special issue, and surveys are usually welcome to let interested researcher get started with the topic.

My main critical points are related to the survey methodology.

First, the selection of the articles. Authors state that they found 1426 publications and that they focused on 71. Authors explain that they selected the papers based on "their quality, their innovative aspect and for the balance in their content between semantic web and IoT". It is very hard to understand how they measured these values. For example, how do the authors judge the quality of the paper? The methodology deserves a more depth explanation in the paper. As an example, this article contains a proper explanation of the methodology: Kitchenham et al., "Systematic literature reviews in software engineering – A systematic literature review", Information and Software Technology, 51(1), but it is very easy to find other papers following similar structures. In my opinion, the selection of the articles is one of the most important parts of this article, since most of the findings are related to this set. For instance, Section 3.1.1 states: "The importance of the enrichment process is underlined by the important number of publications contributing to it: 17 papers in this survey", and Section 3.2 reports: "Discovery is a widely implemented process (10 contributions in this survey)". Without a clear explanation about how the papers were selected, it is hard to accept these results.

Second, the model used to classify the contributions. Authors state that LMU-N covers at least partially 72% of the articles, and situates precisely 53% of the articles. The fact that 28% of the articles do not fit can be explained by the fact that articles on semantic web and IoT have a slightly different focus. In my opinion, this suggests that LMU-N is not a model to capture semantic web contributions to IoT, but it captures something more specific, e.g. semantic web contributions to IoT architectures. However, if this was the scope of the paper, the first part should be reviewed to refine the scope, and the search methodology should be revised as well.

The paper is also very hard to read. There is a huge number of typos, which make hard to focus on the content. Moreover, there are several repetitions that lead on the overall readability of the article.

I report below some detailed comments.

1.1: "The diversity of these application domains is the source of the very high technological heterogeneity in the IoT": is it really the cause? There are a lot of heterogeneity problems also in the same application domains, given by the different types of "things" involved, the different goals, the different protocols.

Section 2
2.1: Section 2.1 sounds more as "Method" than "Motivations".
2.2: What are those architectures that you considered to build LMU-N? You should at least cite them in Section 2.2
2.3: In Sec 2.2 you present the DIKW pyramid, but you do not discuss W (wisdom) at all. What is the reason behind this choice?
2.4: How does Section 2.3.3 (Ontologies for the SWoT) relate to Section 2.3 (Using LMN-U to classify SWoT contributions)?
2.5: It is very unusual to write a whole section (Sec. 2.4) as a pointed list. You may just remove them and use paragraphs instead

Section 3
3.1: Enrichment: as written above, the 17 papers covering this point is not a strong claim. The same holds for the 47% of articles that cannot be situated precisely into LMU-N
3.2: Lowering: while I agree on the importance of this action, I am not very sure that it implies moving downward in the DIKW pyramid. A system that gains the knowledge to decide actions is not losing such knowledge when lowers the action for the actuator.
3.3: What is the relation between control and lowering (w.r.t. LMU-N)? They seem to largely overlap.
3.4: Why is Querying described in "Transport" and not in "Processing content"? Querying is a type of processing, isn't it?
3.5: I disagree on the fact that Aggregation produces "new content instance of the same level in the DIKW hierarchy". Actually, aggregation is one of the most classical ways to move up in this pyramid. We can consider for example censuses, where aggregations over population data create information and knowledge. Moreover, the term aggregation referred to data, information and knowledge usually assume different meanings.
3.6: You state that consistency is not a topic limited to the SWoT, but it looks to me that the same argument holds for most of the points you discuss in Section 3. I agree that "generic consistency mechanisms could be applied to IoT datasets", but the interesting question, in my opinion, should be related to which peculiarities of the IoT domain can be exploited to let such mechanisms perform better in this specific scenario.
3.7: I do not understand what is the point of the last paragraph of Decision Support
3.8: Why did you choose to name two processes of your model with the same name, i.e. Abstraction in Sections 3.1.3 and 3.2.2? It is a bit confusing.
3.9: The identified trends follow are an outcome of the selection of the articles. This section would benefit from a more detailed description of the selection process, as motivated above.

Section 4
4.1: Sec. 4 states: "semantic web contributes to the emergence of the SWoT, but semantic web technologies and principles must also be adapted to meet the constraints of the IoT in order to develop the SWoT". It is not clear to me how these two facts indicate that contributions between semantic web and IoT are not unidirectional. It happens very often that adaptations are required to apply some technologies in some specific domain. As a consequence, it is not evident to me the difference between the contributions described in Sections 3 and 4.

Section 5
5.1: What does "the use of decentralized approaches to LMU-N processes is also crucial to scalability" mean?
5.2: What is the goal of the last four paragraphs of the paper (last paragraph at page 20, paragraphs at page 21)? It is not clear if they are final remarks or future work. Maybe it can be better framed.

The text is full of typos, and a proper revision is required. For example:
- UN (LN, MN) is used in both singular and plural form. A possible solution is to use UN (LN, MN) as singular name, and UNs (LNs, MNs) as plural one.
- All the articles in the bibliography have the title ending with comma followed by double quote, but it should be the opposite (double quote followed by comma)

I report the ones I found in the introduction (the list is too long to report all of them):
deployment of devices and services networks -> deployment of device and service networks
Dublin (IR) -> Dublin (IE)
(JA), that are -> (JA), which are
to monitor their environment -> to monitor their environments
”Things” -> “Things”
2009[3] -> 2009 [3]
manufacturing... -> it is a bit informal to use "..." and "etc.", so they should be avoided in scientific articles. A possible way to rephrase is: "by the IoT, such as environmental metering, [...], and manufacturing."
an application developers should -> "an application developer should" or "application developers should"
diversity[7] -> diversity [7]
feed the Linked Open Data (LOD) -> feed the Linked Open Data (LOD) cloud/project
section 2 defines -> Section 2 defines
On the other hand, section 4 -> when you use "on the other hand", there should be "on the one hand" before

Review #2
By Sebastian Bader submitted on 22/Aug/2017
Major Revision
Review Comment:

The article presents the current developments in the Internet of Things and, in particular, how and which technologies from the semantic web could solve the common interoperability challenge for heterogeneous IoT devices. In order to structure the mentioned works, a hierarchy consisting of three ‘nodes’ (upper, middle, and lower nodes) is introduced. The hierarchy is then used to illustrate IoT related processes and to classify recent works from the semantic web community facing interoperability issues at different levels.

(1) Suitability as introductory text, targeted at researchers, PhD students, or practitioners, to get started on the covered topic.
The paper outlines the contributions of semantic technologies to IoT processes in an extensive and detailed manner which does, at least to my knowledge, not yet exist for recent publications. The categorization scheme in section 3 seems to cover all necessary tasks and assists the reader to reach a detailed impression on the current state of the SWoT. Together with the analysis on already established adjustments of semantic web technologies towards IoT characteristics in section 4 a general overview is reached.

(2) How comprehensive and how balanced is the presentation and coverage.
The authors cover many aspects of the recent literature but miss an objective selection mechanism. Furthermore, standardization initiatives like W3C’s WoT Community Group, Industrial Internet Consortium or Plattform Industrie 4 which already have and further will have significant impact on the development of the IoT are explicitly not considered. Ignoring them must produce an incoherent picture of the sate-of-the-art.
Therefore, many important aspects with potentially high influence are not mentioned.

(3) Readability and clarity of the presentation.
In general, the text is well written and easy to follow. Some comments regarding typos and layout are included below.

(4) Importance of the covered material to the broader Semantic Web community.
Regarding the high possible impact of semantic technologies in this domain a comprehensive summary of the current efforts towards a more interoperable IoT is highly relevant for the semantic web community.

The first chapter gives a brief introduction to IoT, WoT and SWoT thus clarifying the most important relations. Nevertheless, the specific characteristics and challenges are not sufficiently mentioned. ‘Interoperability of heterogeneous devices’ is a very vague problem specification which in fact is further detailed into syntactical and semantical interoperability issues. But a relation to the different characteristics of IoT, WoT and SWoT is missing. The variety of communication protocols at the IoT level are not relevant for the Web of Things (only HTTP) and RDF covers the data format problematic for the Semantic Web of Things. Consequently at least syntactic interoperability is already achieved with basic Semantic Web technologies.
Typos: …devices and services networks.

Section 2
The article switches between presenting the LMU-N concept and the results of the literature review. Even though LMU-N acts as an intuitive pattern to illustrate certain IoT interactions and architectures, the definitions are ambiguous and open to misinterpretations. First, nodes seem to refer to ‘things’ (devices or services) in the Internet of Things. If so it is not clear why an additional terminology is then necessary. Second, the differentiation of Middle Nodes to the others is hard to gasp. If they are mainly defined by the purpose they serve, namely being a gateway for Lower Nodes, then why not using the term ‘gateway’? Third, characteristics like ‘medium processing power’ or ‘restricted memory storage’ vs. ‘limited … storage’ are fuzzy and apply on nearly everything in the domain.
In addition, the nature of Lower Nodes is hard to understand. As they serve as connectors to the real world, Bluetooth connected sensors or actuators should be valid Lower Nodes. But as far as Lower Nodes are described, I assume them to be accessible with Web protocols.
The ambiguous definitions directly lead to difficulties following the author’s argumentation. For example, it is not comprehensible why an Upper Node must not connect to a Lower Node. There are scenarios where this is a reasonable design decision, and it’s definitely in the scope of the IoT.
As ontologies and globally unique vocabularies are one main contribution of the semantic web, section 2.3.3 belongs to section 3 or 4 and should be further extended. Even though published surveys are referenced they play a central role for the SWoT and need more consideration.
2.2.1 Not all nodes of an IoT network are equivalent.
2.2.2 Representing exchanges between nodes as messages flows
2.3.2 …description of nodes characteristics…

Section 3
The relevance of a process type is measured by the number of papers in the reviewed set of papers. But it is not clear why this should reflect a representative selection as an objective selection criteria for regarded papers is missing. In addition, the arrangement of tables 3, 4, and 5 is unfortunate. I would suggest to somehow combine them on one page (at least 3 and 4). In the current setting, table 5 is placed above table 4 which doesn’t help either. Going on, the amount of papers in the ‘unspecified’/’non-LMU’ rows (even some are mentioned more than once) is above 50% for every table. Could that be an indicator that LMU-N might not be a sufficient categorization concept for SWoT?
The process in section 3.1.1 is named ‘enrichment’ while other literature refers to the process as “lifting”. If there is no specific reason I would adjust the notion. Continuing, the ‘visualization’ process somehow does not suit to the other regarded ones. For now, I can not see why it is important for the IoT or how semantic technologies are a crucial enabler.
In the last paragraph of the ‘exposition’ subsection the authors state that only LN need exposition capabilities. Are UN and MN discoverable by default?
Another terminology issue happens in section 3.2.1 where the authors write about ‘non-negotiable criteria ... and negotiable requirements’. ‘Functional’ and ‘non-functional’ might be better suited, especially when referring to service composition. In this paragraph, the topic of (Web) service selection and Node selection get a bit mixed up. Continuing, section 3.2.2 contains a subsection called ‘Abstraction’. This is confusing as the abstraction process is already part of 3.1.3. Also, the mentioned (Composite) Virtual Objects – new nodes created by the composition of existing ones – are not related to LMU-N at all.
At the bottom of page 15 it is stated that ‘contributions dedicated to the same process but at different levels in LMU-N do not expect similar outcomes’. Although it seems trivial that processes in different situations can produce different outcomes a bit more clarification on the statement can improve the reader’s understanding.
3.1. Content-related processes classification…
3.1.1 ...but the transformation process ad-hoc: …
3.1.1 that calls the procedures
3.1.3 The authors also propose several steps to abstraction, …
3.2.1 Similarily, SPARQL queries can be issued…
3.2.1 ...where services descriptions…
3.2.2 The three processes dedicated to homogeneity are focus on the management of this diversity.
3.2.2 Globally, a/the specification process as the lowering of a generic, virtual representation…
3.3 ...nodes constraint the processes they interact with:

Section 4
This section explicitly states some of the major constraints the IoT brings to the SWoT. Yet, I would like to have read them earlier in the article in order to have a better understanding of the specific challenges (computational load, data size) of semantic technologies in the IoT. In contrast, the section on ontology-based data access falls a bit short in comparison to its meaning as it is one of the main advantages the semantic approach provides.

Section 5
The last section repeats the concepts WoT and SWoT and summarizes the mentioned relations to the IoT and the proposed LMU-N. It is worth mentioning that in the middle of the paper, in particular section 3, ‘WoT’ is nearly mentioned but ‘IoT’ quite frequently. Probably at many occasions ‘WoT’ would be more appropriate (e.g. page 14: ‘Therefore, being able to abstract their representation as soon as possible in the IoT hierarchical network makes their management easier for MN.’).
The first paragraph on page 21 refers to combining the IoT with natural language. Even though ontologies may serve as this bridge it is not mentioned anywhere in the text, or I could not relate it to a presented section except to the very last paragraph of the whole paper.
Last, the sentence ‘If simple alignments are not sufficient, complex alignments should also be considered’ should be omitted as it does not provide additional information but creates new questions.